The Performance Optimization Of Data Center Networks In SDN

Explicit Congestion Notification(ECN)is playing a crucial role in the congestion con-trol of Data Center Networks(DCNs).Most DCNs use a single queue scenario in each switch port.However,in the production of DCNs,the industry trend is moving towards one farther queue per-port.Therefore,Multi-Service Multi-Queue Data Centers(MQ-ECN)have been proposed to ignite this service.Having a Multi-Queue(MQ)per-port is harmful,at least for a scheduling policy of ECN.By seeking solutions to the prob-lems mentioned above,we propose a series of schemes.1)ECN~+,the intuition is that if any packet gets marked in the MQ buffer,ECN~+will not consider them in the mark-ing decision of the Output Port Buffer(OPB).Afterward,we propose 2)Deep-RL-a marking decision with Deep Reinforcement Learning(DRL)via per-port for solving the same problem.Thus,we formulate the statement as a DRL problem and use a deep neural network to achieve the agent's best possible policy.Then,we also propose3)Priority-ECN.The intuition of Priority-ECN lies in the fact that if any packet gets marked,it should be prioritized to pass a route quicker than the others to notify the net-work's condition in an early stage.Furthermore,to reach the goal of high throughput without sacrificing the latency,Priority-ECN uses an approach similar to cut payload,which drops the payloads of packets,rather than the metadata,when a queue reaches the threshold.Moreover,we need a short Flow Completion Time(FCT).Thus,to have a short FCT,we need a shallow and deep ECN marking threshold in each switch buffer,re-spectively.To solve the problem,we propose 4)DC-ECN–a first systematic solution to have low latency without sacrificing throughput in Data Centers using ECN.The main point of DC-ECN is a separation of the mice and elephant flows in dual couple queues using machine learning.Then,locate them into the requested queue with min and max ECN threshold in each dual couple buffer.Finally,we have studied the fundamental problem of software-defined networking controller in DCNs,i.e.,when and where to apply reconfigurations.Thus,we propose 5)Mi Fi,which aims to Minimize Flow cost or intuitively average transmission delay,under reconfiguration budget constraints in DCN.Furthermore,we formulate this optimization problem as a constrained Markov Decision Process and propose a set of algorithms to solve it in a scalable manner.We first develop a propagation algorithm to identify the flows which are mostly affected in terms of latency.Then,we set a limitation range for updating them to improve adapt-ability and scalability by updating a less number of flows each time to achieve fast operations.Further,by using Drift-Plus-Penalty in the Lyapunov scheme,we present a heuristic model with the absence of prior knowledge of flow demand and a renewal scheme accompanied by an efficiency guarantee to make the additive optimality gap minimize.Moreover,to the best of our knowledge,Mi Fi is the first paper that stud-ies the range and frequency of flow reconfigurations,which has both theoretical and practical significance in the related area.Steady-state analysis and simulation results demonstrate that ECN~+utilizes the buffer capacity,exactly 30%,DC-ECN achieves21.8%and 16.5%less flow completion time,and Mi Fi outperforms the state-of-the-art algorithms in terms of latency by over 45%while making improvements in adaptability and scalability.